Bounded Badger

[amiller]

Bounded Badger

[lishuai87]

Since the protocol's security relies on eventual delivery of every message, the only way to emulate this in an unreliable network is to store and resend messages indefinitely until a read receipt is obtained.

However, the need to provide unbounded buffers means that the software may end up consuming all the machine’s available disk space or RAM. If messages are dropped when the buffers full up, then the network model no longer fulfills the guarantees assumed by the security analysis, and hence the security proof cannot be guaranteed.

Hi, in a complicated network, we may have many nodes crash (or restart, power-fail) at the same time, the network may randomly drop messages.
In this case, store and resend indefinitely until a read receipt is obtained means we need write (and sync) every message to DB ?
If receiver replied a receipt and then crash, the sender will not resend the message, so we also need write (and sync) every incoming message to DB ?

outgoing message:

VAL
ECHO
READY
BVAL
AUX
CONF
TERM

incoming message:

incoming VAL
incoming ECHO
incoming READY
incoming BVAL
incoming AUX
incoming CONF
incoming TERM

For one round, we need at least 14 writes (and 14 syncs) ? It's a huge burden for real projects.

What if we don't store and resend every message, just let the receiver query missing messages?
In this case, we need store RBC input data, BA input data, and coin data. When node restarts, and it finds missing message, it sends query request to other nodes, then other nodes resend the missing messages.
This only need 2 or 3 writes. But I don't know if this breaks the security guarantee...